ABSTRACT
Body tissues are exposed to a complex mechanical environment, which is perceived by cells and converted to biochemical signals such as ATP release. We performed a meta-analysis of 278 systematically identified studies that investigated mechanically stimulated ATP release (MSAR) to quantify the amounts, kinetics and mechanisms of ATP release under normal and pathological conditions. Mechanically stimulated mammalian cells were shown to release 38.6 [95% confidence interval (CI): 18.2–81.8] amol ATP/cell on average with a characteristic time constant of 32 s (95% CI: 16–66). Analysis of ATP release mechanisms revealed the existence of conserved and tissue-specific release routes. We assessed ATP release in pathophysiological states, and found that ATP release was elevated in inflammation and injury, and attenuated in hereditary (such as cystic fibrosis) and metabolic (such as type II diabetes) conditions. Our study links cell-specific ATP release mechanisms to pathophysiological changes in ATP release and allows ATP release-targeting interventions to be mapped to site-specific effects. This work demonstrates that quantitative synthesis of basic research can generate non-trivial hypotheses and inform evidence-driven translational studies.
INTRODUCTION
The discovery that mechanical stimuli elicit adenosine triphosphate (ATP) release from mammalian cells was made over 25 years ago (Milner et al., 1990), and represented a paradigm shift in how we understand the interaction between the body and the mechanical environment. The body is constantly exposed to a complex combination of shear forces, strains, osmotic stresses and pressures. These forces are perceived at the cellular level and converted to biochemical signals, one of which involves the release of ATP, a high-energy content purine molecule that represents the primary source of cellular energy (Bergman, 1999). It is now clear that extracellular ATP is as an autocrine and paracrine signaling molecule that exerts its effects through purinergic receptors (Burnstock, 2014).
In this study, we systematically reviewed the literature pertaining to mechanically stimulated ATP release (MSAR) from mammalian cells and conducted a meta-analysis on relevant studies. The primary objective was to provide quantitative estimates of the amount of ATP released and kinetics of this release in response to mechanical stimulation. The secondary objective was to extend our analysis to address problems that are otherwise difficult to investigate at the single-study level, including comparing the effects of different mechanical stimuli, as well as consolidating the available intervention- and pathology-related data. In addition to our research objectives, we sought to demonstrate the feasibility and value of conducting meta-analyses at the basic research level to complement existing lines of evidence used to inform translational research.
RESULTS
Overview of relevant studies
We systematically identified 278 studies that directly measured ATP release from mechanically stimulated mammalian cells, 228 of which were included in quantitative synthesis (50 studies were excluded due to inadequate reporting) (Fig. 1A). In these studies, 9 unique models of mechanical stimulation, including osmotic pressure, fluid shear stress (FSS), strain and compression, were used to stimulate ATP release from mammalian cells (Fig. 1B). Cells derived from 12 different organ systems were studied, most commonly originating from the urinary and musculoskeletal systems (Fig. 1C). Most studies measured ATP using the luciferin-luciferase bioluminescence assay (255/278), while others used high-performance liquid chromatography (11/278), hexokinase-based enzymatic assays (5/278), surface-bound luciferase probes (3/278), microelectrodes (3/278) and cell-based biosensors (1/278). We tracked the discovery of non-lytic ATP release to Milner et al., who described the phenomenon in FSS- stimulated endothelial cells in 1990 (Milner et al., 1990) (Fig. 1D). The physiological relevance of MSAR was soon after realized and widespread interest surged around the 2000s. It is now clear that MSAR is physiologically involved in virtually every organ system and plays a role in facilitating responses to mechanical perturbations that demand immediate physiological feedback (Fig. 1E). To date, the general physiological relevance has been established, and current work in the field is focused on understanding the mechanistic and potential therapeutic relevance of MSAR.
Overview of studies in the field of MSAR. (A) Simplified PRISMA diagram of the information flow throughout systematic screening process. N, number of studies. (B) Schematics of mechanical stimuli studied. Pie charts show the proportion of studies focused on specified mechanical stimulus. Strain includes tissue distension and substrate stretch; vibrational includes ultrasonication and substrate vibration. (C) Pie chart of relative frequency of organ systems studied. (D) Milestones in the field of MSAR. The timeline shows key publications or notable discoveries; main routes of ATP release identified are indicated in bold text; the bar graph (green) shows the number of relevant publications per year; asterisks indicate key contributions published after systematic search. (E) Overview of the physiological role of MSAR in different tissues. CALHM1, calcium homeostasis modulator 1; CFTR, cystic fibrosis transmembrane conductance regulator; ECM, extracellular matrix; FSS, fluid shear stress; g, gravitational force; PGT, prostaglandin transporter; PPi, pyrophosphate; RBC, red blood cell; VDAC, voltage-dependent anion channel; VRAC, volume-regulated anion channel; VSCC, voltage-sensitive calcium channel; VNUT, vesicular nucleotide transport.
MSAR is a conserved phenomenon in mammalian cells
We quantified ATP release as the amount of ATP liberated into the extracellular space after mechanical stimulation (Fig. 2A). Using a random effects model, we estimated that mechanically stimulated mammalian cells released 38.6 [95% confidence interval (CI) 18.2–81.8] amol ATP/cell over a basal extracellular amount of 8.1 (95% CI: 3.9–16.6) amol ATP/cell. The total intracellular ATP content was estimated to be 5.0 (95% CI: 2.6–9.5) fmol ATP/cell in nucleated cells and 0.14 (95% CI: 0.12–0.18) fmol ATP/cell in red blood cells (RBCs) (Fig. 2B; Table S1). ATP release estimates were considerably inconsistent between studies, ranging over 10 orders of magnitude (Fig. S1, Table S1). We found that 80% of the heterogeneity was explained by a correlation between basal ATP (Abase) and ATP release (Amech) estimates that arose only between studies but was absent within studies (Fig. 2C). Furthermore, we found that heterogeneity in basal ATP was not explained by differences in methodologies (Fig. 2D), suggesting that the primary source of heterogeneity was inconsistent ATP calibrations between studies. Estimates of relative ATP release were more consistent between studies, and were unaffected by the overall quality of the study, although some publication bias was evident (Fig. S1). We found that every cell type investigated responded to mechanical stimulation by releasing ATP, resulting in a 4.3-fold (95% CI: 3.8–4.8) increase in extracellular ATP above baseline (Fig. 2E). Remarkably, the amount of ATP released was generally robust to variations in experimental methodology and biological characteristics (Fig. S2, Table S2). In particular, different cell types, cells from different species, embryonic origin and most organ systems released similar amounts of ATP upon mechanical stimulation. Among the factors that did influence ATP release, we found that neutralization of ATP degradation or application of cyclic mechanical stimuli were associated with relatively lower ATP release (Fig. S2A), and polarized epithelia released significantly higher amounts of ATP from the apical surface compared with the basolateral membrane (Fig. S2B). We conclude that MSAR is a conserved phenomenon in mammalian cells, and provide robust quantitative estimates for the absolute and relative amounts of ATP released upon mechanical stimulation.
Quantitative analysis of MSAR from mammalian cells. (A) Schematic representation of intracellular (IC) ATP content (Acell, blue), basal extracellular (EC) ATP (Abase, gray) and ATP released by mechanical stimulation (Amech, red). The number of datasets is shown in parentheses. (B) Forest plot of log-transformed study-level outcomes ±95% confidence intervals (CI). The shaded bands indicate 95% CI for overall 
(blue), 
(gray) and 
(red). Study-level sample sizes are in parentheses. (C) Meta-regression analysis of the relationship between 
and 
. Black lines indicate meta-regression; dark/light-red bands indicate 95% CI and prediction intervals, respectively; the gray band indicates 
95% CI; red lines indicate within-study regression. R2explained, heterogeneity explained by regression; I2, variance due to heterogeneity; βinterstudy, slope ±95% of the between-study relationship; βintrastudy, average slope ±95% of the within-study relationship. (D) Effects of experimental factors on 
. Black markers/lines indicate subgroup estimates ±95% CI; the red band indicates overall 
95% CI. (E) Forest plot of study-level 95% CI for relative ATP release 
. The shaded band indicates overall 95% CI for 
; the dashed black line is the no ATP released reference. H2, heterogeneity statistic; N, number of included datasets. Detailed overall statistics are provided in Tables S1 and S2.
The amount of ATP release is proportional to the magnitude of mechanical stimulation
We next investigated the relationship between the type of mechanical stimulation and ATP release. We found that similar amounts of ATP were released in response to all studied mechanical stimuli (Fig. 3A; Table S3). Only locally applied membrane deformations resulted in lower ATP release compared with other forms of stimulation. We used a combination of meta-regression and within-study regression to investigate the relationship between the magnitude of mechanical stimulation and the amount of ATP released (Fig. 3B–G; Table S3). Within studies, ATP release was consistently proportional to the magnitude of mechanical stimulation, whereas between studies this relationship was less pronounced. Nevertheless, the directions of association were consistently positive for all types of mechanical stimulation (Fig. 3H). Thus, despite evident aggregation bias, we conclude that the amount of ATP released is governed by the magnitude of mechanical stimulation.
Relationship between mechanical stimuli and ATP release. (A) Relative amount of ATP (Rmech) released in response to different mechanical stimuli. The number of datasets is shown in parentheses. Black circles indicate subgroup estimates; horizontal black lines indicate ±95% CI; horizontal red lines indicate ±Bonferroni-adjusted 95% CI; the red band indicates 95% CI for overall Rmech; the dashed black line is the no ATP released reference. (B–G) Meta-regression analysis of the relationship between the magnitude of mechanical stimulation and 
following stimulation by strain (B), compression (C), hypotonic pressure (D), hypertonic pressure (E), FSS (F) or deformation of RBCs (G). Black lines indicate meta-regression; dark/light-red bands indicate 95% confidence and prediction intervals; gray bands indicate 95% CI of 
for the indicated stimulus; red lines indicate intrastudy regression. Insets are schematics of mechanical stimuli. (H) Comparison of intrastudy (βintra, red) and interstudy (βinter, black) regression slopes ±95% CI. Red bands indicate 95% CI overlap; dashed lines are the no relationship reference. Detailed statistics are provided in Table S3.
ATP release kinetics are stimulus dependent
To quantify the kinetics of ATP release, we fitted study-level time-series data to a sigmoidal model using a Monte Carlo fitting procedure and estimated the time to half-maximal (half-max) ATP release thalf (Fig. 4A). Meta-analysis demonstrated that the characteristic half-max ATP release time was 101 s (95% CI: 83–117 s) (Fig. 4B). ATP release kinetic estimates were moderately heterogeneous with no discernable publication bias, and they were not influenced by study quality (Fig. S3). However, we found that real-time online recordings of ATP release yielded significantly faster release kinetics of 32 s (95% CI: 16–66 s) compared with offline measurement methods, which yielded slower kinetics of 136 s (95% CI: 117–159 s) (Fig. 4C,D; Table S4). We stratified the data by recording method prior to further analysis. ATP release kinetics were robust to differences in all other experimental factors but not to biological factors (Fig. S4, Table S4). ATP release was relatively slower in certain species (guinea pigs, porcine, rabbits) and cells from different embryonic origins (ectoderm) and organ systems (sensory). Importantly, the kinetics of ATP release depended on the type of mechanical stimulus applied. FSS evoked significantly faster ATP release compared with osmotic pressures (Fig. 4E,F; Table S4). Thus, systematic large-scale data synthesis allowed us to identify a previously unknown relationship between ATP release kinetics and the type of mechanical stimulation, suggesting that it needs to be further explored experimentally.
Kinetics of ATP release. (A) Study-level ATP release kinetics (black circles) were fit to sigmoidal function using the Monte Carlo method and characteristic time to half-max release (thalf, red) with standard error was estimated. The dashed black line indicates the maximal ATP released (Amax); the red band indicates sigmoidal function 95% CI. (B) Forest plot of log-transformed study-level 95% CIs for thalf 
. The shaded band indicates the 95% CI for overall 
. N, number of datasets; H2, heterogeneity statistic. (C,D) Subgroup estimates of thalf±95% CI (C) and 95% CI temporal profiles of ATP release (D) recorded using online or offline methods. The red band indicates the 95% CI for overall thalf. (E,F) thalf stratified by recording method for different mechanical stimuli. (E) Subgroup thalf estimates. Green, offline recording; yellow, online recording; horizontal black lines, ±95% CI; horizontal red lines, ±Bonferroni-adjusted 95% CI; bands/diamonds, overall thalf ±95% CI. (F) Comparison of FSS- (blue) and osmotic pressure- (red) stimulated ATP release 95% CI temporal profiles when recorded online (top) or offline (bottom). Detailed statistics are in Table S4.
Mechanisms of MSAR
We next synthesized the data pertaining to the mechanisms of MSAR. Pharmacological and genetic interventions were grouped by common molecular targets (Table S5) and their relative inhibitory effects were quantified for each cell type (Fig. 5; Table S6). The five main direct routes of MSAR that have been identified in mammalian cells are vesicles (Bodin and Burnstock, 2001; Sathe et al., 2011), pannexins (Locovei et al., 2006), connexins (Graff et al., 2000), volume-regulated anion channels (VRACs) (Pedersen et al., 1999; Qiu et al., 2014; Voss et al., 2014) and maxi-anion channels (MACs) (Sabirov et al., 2001; Sabirov et al., 2017). Vesicles were involved in 70% (14/20) of studied cell types from all organ systems except for the integumentary (i.e. keratinocytes, mammary epithelial cells) (Fig. 5A). Pannexins were the second most common route of ATP release, involved in 58% (15/26) of all studied cell types. Pannexin-mediated ATP release occurred in hematopoietic cells, astrocytes and most epithelial cells (including digestive, airway and ocular). Connexins, VRACs and MACs were implicated in ATP release in 35% (7/20), 36% (4/11) and 23% (3/13) of all cells studied, respectively. Connexin-mediated ATP release was highly prevalent in ocular cell types and in the kidney, but was absent in airway epithelia and smooth muscle. VRACs exhibited no discernable tissue-specific pattern of involvement, while MACs were almost exclusively implicated in ATP release from integumentary cells. This data-driven summary of MSAR mechanisms allowed us to establish tissue-level generalizations and suggest the existence of common (vesicular) and tissue-specific (pannexin, connexins, MAC) routes of ATP release.
Mechanisms of MSAR. (A) Involvement of the most commonly studied MSAR mechanisms (columns) for each cell type (rows). Stacked bars indicate the relative frequencies of involvement for each mechanism. The number of unique cell types studied per mechanism is indicated. Dark green/red indicates findings replicated by a separate study/method; light green/red indicates findings that are not replicated. (B) Percentage of cell types in which implicated mechanisms co-occurred. Results shown are for mechanism pairs studied in >2 cell types. Asterisks indicate pairwise co-occurrences confounded by common pharmacological interventions. (C) Relative contributions of main ATP release routes to ATP release in cells stimulated by FSS, osmotic pressure or strain, approximated by inhibitory effect (%)±s.e. of interventions. **P<0.01 and ***P<0.001 indicate significant contributions as assessed by Student's t-test, compared with reference value 0 (i.e. no contribution to ATP release). Numbers of datasets are indicated. Intervention data used to construct matrices are provided in Tables S5 and S6. B.M., bone marrow; [Ca2+]e, extracellular calcium; [Ca2+]i, intracellular calcium; CFTR, cystic fibrosis transmembrane conductance regulator; endo, endothelial cells; epi, epithelial cells; fibro., fibroblast; FSS, fluid shear stress; Glomer., glomerular; kin., kinase; MSC, mesenchymal stem cells; mTs, microtubules; Pul. Art., pulmonary artery; S.M., smooth muscle; Trab. mesh., trabecular meshwork cells; Um., umbilical; VRAC, volume-regulated anion channel.
We conducted a co-occurrence analysis to screen for possibly co-dependent pathways of ATP release (Fig. 5B). The main routes of ATP release (i.e. vesicular, pannexin, connexin, VRAC and MAC) co-occurred pairwise in 27–67% of studied cells, except for vesicle- and MAC-mediated ATP release routes, which were mutually exclusive (co-occurred in 0 of the 4 cell types studied; 0/4). We investigated whether the main routes of ATP release contributed to the total amount of ATP released in an additive manner by accounting for the relative contributions of each route (Table 1). In certain cell types, the total amount of ATP released was accounted for by the additive contribution of the main routes of ATP release. However, in cardiomyocytes, ligament cells, pancreatic and renal epithelia, and urothelial cells, the additive contributions exceeded the amount of ATP released. Although there is a possibility that pharmacological non-specificity contributed to this effect, genetic methods were in agreement. For instance, small interfering RNA (siRNA)-mediated knockdown of Cx43 (also known as GJA1) and Panx1 in ligament cells inhibited ATP release by 73.1% (95% CI: 68.7–77.4) (Luckprom et al., 2011) and 73.6% (95% CI: 69.2–78.0) (Kanjanamekanant et al., 2014), respectively. These findings suggest the existence of a cell type-specific synergistic interaction between ATP release routes previously thought to be functionally independent.
Total contribution of the main routes of ATP release
We investigated the association between the main ATP release routes and other implicated regulatory and auxiliary pathways. Vesicular ATP release consistently coincided with the involvement of intracellular calcium ([Ca2+]i, 8/8 cell types studied), Rho kinases (3/3) and microtubules (4/4). Rho kinases (3/3) and microtubules (3/3) were also involved in pannexin-mediated ATP release. No pathway consistently co-occurred with connexin-mediated ATP release. VRAC-mediated ATP release coincided with the involvement of P2X7 receptors (5/5), but never with ATP synthase (0/3), and MAC-mediated ATP release co-occurred with ATP synthase-related ATP release (3/3), but never with [Ca2+]i-dependent ATP release (0/3). Thus, we have identified conserved cell-independent signaling patterns that warrant further experimental investigation.
We next explored whether the mechanisms of ATP release depend on the type of mechanical stimulation (Fig. 5C). We examined the contribution of the main routes of ATP release in response to mechanical stimuli with sufficiently large datasets (FSS, osmotic pressure and strain). All 5 routes of ATP release were implicated in osmotic pressure- and strain-induced ATP release with the similar patterns of contribution. In contrast, VRAC and possibly pannexins were not involved in FSS-induced ATP release. Thus, consistent with our findings that FSS- and osmotic pressure-induced ATP release exhibit distinct release kinetics, these data suggest that distinct mechanisms may contribute to stretch- and shear-related responses.
MSAR in pathologies
Aberrant ATP release has been implicated in multiple pathophysiological conditions (Fig. 6A). Inflammation- and injury-related conditions were associated with elevated levels of MSAR, compared with unaffected controls (Fig. 6B; Table S7). On the other hand, attenuated ATP release was observed in hereditary conditions, such as cystic fibrosis and xerocytosis, and in type II diabetes and primary pulmonary hypertension (Fig. 6C; Table S7). In some pathologies, the magnitude or direction of changes in MSAR were specific to the cell or stimulus. In cystic fibrosis, RBCs and pancreatic epithelia released less ATP upon stimulation, while airway epithelia and astrocytes were unaffected. Acute hypoxia potentiated ATP release from endothelial cells, while chronically hypoxic cells released less ATP (Fig. 6D; Table S7). In polycystic kidney disease (PKD), renal epithelia released less ATP upon FSS stimulation in both autosomal dominant and recessive cases, whereas osmotic swelling-induced ATP release was unaffected in autosomal recessive and potentiated in autosomal dominant PKD (Fig. 6D; Table S7), again suggesting distinct cell processing of shear- and stretch-related stimuli. Thus, we demonstrate the presence of specific patterns of MSAR alteration with respect to the type of pathology, affected cells and applied stimulus, and suggest that inhibitory and stimulatory interventions targeting MSAR can be of therapeutic interest, but need to be applied with caution. Our study maps the potential anatomical sites and situations in which targeting MSAR could have (patho-)physiological consequences.
Pathological ATP release. (A) Pie chart of relative frequency of clinically relevant conditions studied or implicated in aberrant MSAR (61 studies in total). Blue, increased ATP release; red, decreased ATP release; gray, implicated but not measured; purple, heterogeneous effects; light shade, reported by a single study; dark shade, replicated findings. (B–D) Forest plots of relative effect (%) of pathologies (specified at the top left) on MSAR from cells (specified on the right) compared with unaffected controls. Black markers/lines indicate study-level relative effects ±95% CI; bands indicate 95% CI of overall pathology-specific effect. Marker sizes are proportional to study-level sample sizes; numbers of independent studies for each pathology are in parentheses. Black dashed lines are the no effect reference. (B) Pathologies associated with increased MSAR. (C) Pathologies associated with decreased MSAR. (D) Pathologies that had heterogeneous or no effect on MSAR. ADPKD, autosomal dominant PKD; aq. hum., aqueous humour; ARPKD, autosomal recessive PKD; astro., astrocytes; diabetes, type II diabetes; FSS, fluid shear stress; panc. epi, pancreatic epithelia; PKD, polycystic kidney disease; PPH, primary pulmonary hypertension; RBC, red blood cell; resp. epi, respiratory (airway) epithelia; SCI, spinal cord injury.
DISCUSSION
Overview
We have conducted a systematic large-scale data synthesis to quantitatively characterize the amounts, kinetics and mechanisms of MSAR under normal and pathological conditions. From 228 systematically selected studies, we extracted 123 estimates of absolute and 212 estimates of relative amount of ATP released, 74 kinetic time-series, 592 pharmacological and 89 genetic intervention outcomes, and 51 pathophysiological comparisons. Using a meta-analytic approach, we have established that mechanically stimulated mammalian cells release 38.6 (95% CI: 18.2–81.8) amol ATP/cell, with a characteristic time constant of 32 s (95% CI: 16–66) measured using real-time recording methods. We have found that MSAR is a universally conserved phenomenon in mammalian cells, and that cells from different species, embryonic origin and most organ systems release similar amounts of ATP when mechanically stimulated. Our data-driven summary of MSAR mechanisms allowed us to infer tissue-level generalizations that suggest the existence of common and tissue-specific routes of ATP release, and to identify conserved cell type-independent signaling patterns. We have found that inflammation and injury were associated with increased MSAR, whereas hereditary and metabolic conditions resulted in attenuated ATP release. Importantly, several lines of evidence including (1) differences in release kinetics, (2) implicated mechanisms and (3) pathophysiological effects in PKD, suggest that cells can discriminate between stretch- and shear-related forces. Thus, consolidating and quantifying over 25 years of basic research data generated in 64 unique cell types derived from 12 organ systems and stimulated by 9 distinct force applications allowed us to generate novel testable hypotheses, and provide evidence-driven recommendations for translational studies.
Study limitations
The studies included in this meta-analysis were highly heterogeneous; however, this was expected due to the higher methodological variability in exploratory basic science studies (Bradbury and Plückthun, 2015; Fontoura-Andrade et al., 2017; Soehnlein and Silvestre-Roig, 2017). Importantly, accounting for interstudy differences in ATP calibration and recording methods allowed us to dramatically reduce heterogeneity. We found minimal evidence of publication bias. We also demonstrated that the quality of the studies did not significantly affect study-level outcomes, despite quality scores varying substantially across studies. It is known that subgroup analyses come at the expense of lower statistical power; however, the large number of available datasets permitted statistically powered analysis for numerous secondary outcomes (Jackson and Turner, 2017). Analysis of MSAR mechanisms was limited by overlapping and off-target effects of many inhibitors, as well as lack of within-study inhibitor validation and over-reliance on assumed pharmacological targets. We minimized false-positive outcomes by applying random effects meta-analytic models and considering Bonferroni adjustments for multiple comparison analyses (Glass, 1986). There remains a distinct possibility of false-negative outcomes due to limited sample sizes in some subgroups, aggregation bias and heterogeneity (Higgins and Thompson, 2002).
Quantitative characterization
We estimated that mechanically stimulated mammalian cells release 38.6 (95% CI: 18.2–81.8) amol ATP/cell, resulting in a 4.3-fold (95% CI: 3.8–4.8) increase in ATP above basal levels of 8.1 (95% CI: 3.9–16.6) amol ATP/cell. Intracellular ATP content was estimated to be 3 orders of magnitude higher than basal ATP levels in nucleated cells, 5.0 (95% CI: 2.6–9.5) fmol ATP/cell, and 2 orders of magnitude higher in RBCs, 0.14 (95% CI: 0.12, 0.18) fmol ATP/RBC. Study-level estimates of the absolute amount of ATP released ranged over 10 orders of magnitude, with 5 studies reporting more ATP release than can be contained within a cell, suggesting that more caution must be taken when performing ATP calibrations and measurements, and that basal, released and total ATP content should be reported to obtain relative measures. The characteristic time to half-max ATP release was strongly influenced by the recording method, yielding almost 4 times faster estimates by real-time recordings compared with estimates acquired by bulk sampling and offline measurement. This difference can potentially be explained by the diffusion time required to equilibrate the concentration within the volume of the culture medium. In this regard, the volume into which ATP is released is an important (but not routinely controlled for) determinant of the effective ATP concentration available for autocrine and paracrine signaling, including ATP-regulated ATP release (Bodin and Burnstock, 1996; Dillon et al., 2013). We found that less ATP is released in response to cyclic stimulation; however, no studies reported the kinetics of ATP release in response to repeated or cyclical stimulations, even though physiological stimuli are often cyclical (Burr et al., 1996; Eyckmans et al., 2011; Fritton et al., 2000). Of interest, studies in which cell injury was assessed and detected or intentionally induced reported a tendency for higher ATP release. However, the amount of injury-related ATP release never reached amounts expected following cell destruction and was not statistically different from osmotic pressure- or FSS-induced ATP release. Thus, quantitative synthesis of basic science studies employing diverse approaches with complex endpoints is feasible and has allowed us to identify methodological variations of consequence.
Dependence on mechanical stimulus
We compared ATP release induced by 9 different types of mechanical stimulation, including stretch-related stimuli, such as substrate strain, osmotic pressure and tissue distention, and FSS and local membrane deformation. Although all types of mechanical stimulation resulted in the release of comparable amounts of ATP, we found significant differences in the kinetics of ATP release, which were much faster in response to FSS- compared with stretch-related stimuli. In addition, we found that VRAC and possibly pannexins were involved in mediating swell- and strain-induced, but not shear-induced, responses. Finally, MSAR from PKD-afflicted renal epithelia was differentially sensitive to FSS and hypotonic swelling. These results strongly suggest that mammalian cells can discriminate between different types of mechanical stimuli. Theoretical models have previously demonstrated that shear stresses induce more cell membrane deformation than stretch-related stimuli (Lynch and Fischbach, 2014; McGarry et al., 2005). Direct comparison of mechanisms involved in hypotonic pressure- and strain-induced ATP release has demonstrated that these stretch-related stimuli recruit common ATP release pathways (Li et al., 2011); however, no study has directly compared shear- and stretch-related ATP release. Thus, we recommend a direct comparison between shear- and stretch-induced cell deformation, and ATP release to be investigated in future work.
Intervention studies
We quantified the effects of pharmacological and genetic intervention for 681 combinations of cell type, mechanical stimulation and interventions. From the five main routes of ATP release [vesicular (Bodin and Burnstock, 2001; Sathe et al., 2011), pannexin (Locovei et al., 2006), connexin (Graff et al., 2000), VRAC (Pedersen et al., 1999; Qiu et al., 2014; Voss et al., 2014) and MAC (Sabirov et al., 2001; Sabirov et al., 2017)], at least 2, and often 3, were consistently implicated in the same cell type, with the exception of keratinocytes, in which all 5 pathways were studied, but only MAC was found to mediate ATP release. The involvement of multiple independent release mechanisms may confer a redundancy that ensures that cellular ATP release is robust. Alternatively, it is possible that different routes work collaboratively within the same pathway. Of interest, we demonstrated a lack of additivity in the relative contributions of main release routes in certain cell types. We found that different ATP release routes shared common intracellular signaling pathways. In particular, Rho kinases and microtubules always co-occurred with vesicular and pannexin pathways, and [Ca2+]i co-occurred with all 5 routes of ATP release in 33–100% of cases. Thus, it can be hypothesized that different routes of release are functionally independent but are regulated by common intracellular signaling. At this time, this hypothesis is difficult to test experimentally, partly because pharmacological agents used to study mechanisms of MSAR suffer from extensively overlapping antagonistic profiles (Azorin et al., 2011; Li et al., 2010; Liu et al., 2008; Sauer et al., 2000; Wang et al., 2005). Targeted genetic studies (e.g. siRNA, CRISPR, animal models) are needed to further our understanding of MSAR. To date, genetic interventions have been used to study the involvement of vesicle-related vesicular nucleotide transporter (VNUT; also known as SLC17A9) (Sathe et al., 2011; Sawada et al., 2008), pannexin 1 (Panx1) (Bao et al., 2004; Beckel et al., 2014; Kanjanamekanant et al., 2014; Lu et al., 2012; Seminario-Vidal et al., 2011; Woehrle et al., 2010), pannexin 2 (Panx2) (Oishi et al., 2012), connexin 40 (Cx40; also known as GJA5) (Toma et al., 2008), connexin 43 (Cx43) (Chi et al., 2014; Genetos et al., 2007; Lu et al., 2012; Luckprom et al., 2011) and connexin 45 (Cx45; also known as GJC1) (Lu et al., 2012). Recently, SWELL1 (also known as LRRC8A) was identified as the pore component of the VRAC complex (Qiu et al., 2014; Syeda et al., 2016; Voss et al., 2014), and the prostaglandin transporter PGT (encoded by Slco2a1) was recognized as the MAC (Sabirov et al., 2017). Of interest, Sana-Ur-Rehman et al. (2017) and Workman et al. (2017) have recently reported the calcium homoeostatic modulator 1 (CALHM1) as a novel mediator of MSAR in nasal epithelia and the urothelium; however, it remains unclear whether CALHM1 is a direct or indirect conduit of ATP release. A comprehensive review of recent evidence supporting the role of CALHM1 in ATP release can be found elsewhere (Taruno, 2018). Nonetheless, now that the molecular identity of each main route of ATP release has been identified, genetic studies to resolve ATP release mechanisms are feasible. Thus, systematic analysis of prior data allowed us to pinpoint specific mechanistic features that warrant further experimental investigation, such as differential involvement of VRAC in FSS- and osmotic pressure-induced ATP release, and to suggest that certain ATP release mechanisms are cell type and stimulation type dependent.
Therapeutic potential
Our systematic assessment of MSAR involvement in different pathologies included data from 10 cell types/tissues from 11 pathological conditions. We have found that inflammation and injury coincided with higher ATP release from epithelial cells, which might contribute to pain commonly present in these conditions (Butrick et al., 2010; Docherty et al., 2011; Taweel and Seyam, 2015; Weinreb et al., 2014). In contrast, in hereditary and metabolic conditions, lower ATP release from RBCs was consistently reported. Thus, both MSAR inhibitory and stimulatory interventions can be of therapeutic interest. The downstream actions of MSAR are mediated by 15 members of the purinergic (P2) receptor family (Burnstock, 2014), which have been identified as valuable therapeutic targets for treatment of pain, inflammation, spinal cord injury and bladder dysfunction (North and Jarvis, 2013). There are several advantages of targeting MSAR over the P2 receptor network. First, the impact of disproportionally targeting single P2 receptors has poorly understood implications for signaling by the entire P2 receptor network. Instead, manipulating MSAR allows proportional reduction or increase in the stimulation of all P2 receptors. Second, many of the drugs used to inhibit MSAR, including mefloquine (Lee et al., 2017), carbenoxolone (Doll et al., 1965), probenecid (Li et al., 2016), flufenamic acid (Flemming and Jones, 2015), glybenclamide (Sola et al., 2015) and clodronate (Ghinoi and Brandi, 2002) [recently demonstrated to potently inhibit VNUT (Kato et al., 2017)] are already used in clinic. Although these drugs are relatively non-specific, strategies to therapeutically re-purpose them for diseases with aberrant MSAR may be considered. As for any potential therapy, unintended drug effects need to be taken into account. In this regard, our systematic approach allowed the identification of cell- and stimulus-specific effects of various pathologies. Thus, comprehensive assessment of cell type-specific mechanisms of ATP release, considered together with known pathophysiological changes, can be used to map site-specific effects of therapeutic MSAR interventions, and predict any unintended (patho)physiological consequences.
MATERIALS AND METHODS
Software
Endnote X7 (Thomson Reuters) was used to manage references, the MetaLab meta-analysis toolbox in MATLAB (MathWorks) (N.M. and S.V.K., unpublished) was used for data extraction and analysis, Excel 2016 (Microsoft) was used for data storage and CorelDRAW X8 (Corel) was used for figure preparation.
Search strategy and inclusion criteria
A medical librarian (M.M.) prepared the systematic search strategy. The search strategy was constructed around the key terms ‘ATP’ and ‘mechanical stimulation’. Search terms were validated by ensuring the search retrieved a selection of articles, representative of relevant works. Medline, Embase, Biosis and SCOPUS databases were searched on 12 January 2017, using search terms listed in Table S8, and articles were exported to Endnote. Reviews, books, letters, editorials and conference proceedings were excluded, language was restricted to English and there was no restriction on date of publication. Studies in which mammalian cells were mechanically stimulated and ATP release was assayed were included. No restrictions were imposed on experimental setup. Abstracts were screened independently by two reviewers (N.M. and A.M.). Full-text screens were conducted to confirm eligibility. Differences between two reviewers were resolved through discussion and consensus. The complete list of systematically identified studies is in Table S9.
Data extraction
Study characteristics and data were extracted by a single non-blinded reviewer (N.M.) and independently verified by another reviewer (A.M.) to minimize user-related error. For all studies, study design and biological characteristics were recorded (Table S10). For experiments using a perfusion-based sampling method, perfusion intervals and volumes were extracted to estimate cumulative ATP release. For experiments reporting outcomes in units other than moles ATP per cell, relevant conversion parameters were extracted. For intervention/pathology studies, we collected information on pharmacological agents, genetic targets or pathological states. MetaLab data extraction module was used to facilitate graphical data extraction (N.M. and S.V.K., unpublished). For each dataset, we extracted baseline ATP levels and mechanically stimulated ATP along with measures of variance. In some studies, multiple datasets were extracted if reported. For temporal recordings of ATP release, entire time series were extracted and used to estimate the time to half-max ATP release for analysis of ATP release kinetics. From these time series, the maximal amount of ATP release was also used for analysis of amount of ATP released. For intervention or pathology studies, basal ATP and amount of ATP released in the absence and presence of the intervention/pathology was extracted for subsequent calculation of the intervention inhibitory effect (%) or relative effect (%) of pathology. Sample sizes and type of variance measures (standard error, standard deviation) used were recorded. When variance measure was unclear, error was extracted as standard error, lending to more conservative estimates. When sample size was unclear but there was indication of multiple trials, we set the sample size to 3. When a range of sample sizes was reported, the smallest value was extracted.
Standardization of measures
ATP was commonly reported as an amount or concentration of ATP released per cell(s). Accordingly, conversion factors were estimated for each study to express ATP amount in terms of moles per cell. In cases in which certain conversion parameters were not reported, assumptions were made according to the table reported in Table S11. Cell-related parameters were estimated using the BioNumbers database (Milo et al., 2010), and were compared with our in-laboratory experience and deemed appropriate. Volume assumptions for culture dishes were made per manufacturer-recommended volumes. For each type of mechanical stimulation, if enough information was provided, magnitude of stimulus was converted to a common unit. Experiments that applied osmotic pressure commonly reported changes in osmolarity as a percentage of basal osmolarity (isotonic). Because basal osmolarity varied across studies, we expressed the magnitude of osmotic stress in terms of change in mOsm/KgH2O (δ mOsm/l).
Quality assessment
Quality of studies was assessed according to an 8-item quality checklist: publication in peer-reviewed journal, control of temperature, control of sample degradation, cell viability checked, mechanical stimulus regime reported, ATP calibration, negative control and statement of potential conflicts of interest reported. Outcomes were stratified by aggregate quality score to determine the influence of study quality on reported results.
Study-level outcomes
Five outcomes were synthesized in this study: (1) absolute ATP released (Amech), (2) relative ATP released above baseline (Rmech), (3) time to half-max ATP release (thalf), (4) effect of pharmacological/genetic intervention on ATP release (% inhibition) and (5) effect of pathology on ATP release (relative effect, %).
Study-level absolute ATP release
The amount of ATP released was calculated as the difference in extracellular ATP before and after mechanical stimulation:
(1)where Amech was mechanically induced ATP release, AEC was total extracellular ATP following mechanical stimulation and Abase was basal extracellular ATP.
Study-level relative ATP release
The relative amount of ATP released above baseline Rmech was computed as:
(2)
Estimation of cumulative ATP release
Extracted ATP release time series were reported as cumulative ATP release or rates of ATP release. Release rates data were converted to cumulative ATP release over the reported period following mechanical stimulation:
(3)
where a was the cellular release rate (units ATP per unit time per cell), A(t) was the cumulative extracellular ATP, b was the reported time period and t was the time.
Study-level ATP release kinetics
Study-level ATP release time series recordings with at least 4 time points were fit to a sigmoidal function (Eqn 4) using a Monte Carlo error propagation method, and the characteristic time to half-max release thalf with standard error was estimated. Recordings that extended beyond the time of maximal ATP release were truncated to restrict the fitting of the sigmoid curve to the relevant period over which ATP was released.
(4)
where Amax was maximal ATP release, A(t) was cumulative ATP release with respect to time, n was coefficient related to the slope of activation, t was time and thalf was time to half-max release. The Monte Carlo fitting method allowed us to propagate study-level variances (uncertainty in the model inputs) to the uncertainty in the model parameter estimates (Cox et al., 2003). This approach assumes that study-level estimates are normally distributed, enabling pseudo random observations to be sampled from a distribution defined by the study-level means and standard deviations. The pseudo random observations were then averaged to obtain a Monte Carlo estimate for each time point and the sigmoidal model was fit to the Monte Carlo estimates using the least-squared method. This procedure of pseudo-random sampling and model fitting was iterated 1000 times, and the distribution of 1000 thalf estimates was used to compute a mean and variance that was then used as the study-level estimate of thalf. Model goodness of fit was assessed quantitatively by R2 and visually verified.
Study-level percentage inhibition
For intervention studies, the effect of genetic and pharmacological interventions was computed as percentage inhibition, compared with unaffected control of ATP release:
(5)
and 
were computed the same as Amech (i.e. basal ATP release was accounted for). Negative values of percentage inhibition outcomes indicate increased ATP release, and positive values indicate decreased ATP release.
Study-level relative effects
For pathophysiological conditions, the relative effect size was calculated as follows:
(6)where 
was ATP released from the pathologically affected cell/tissue type and 
was ATP released from healthy control samples.
Quantitative synthesis
For each study i, study-level outcomes θi and standard errors se(θi) were estimated. θi is a general study-level outcome used to represent Amech, Rmech, thalf, inhibition (%) or effect (%).
Data transformation and normalization
Prior to synthesis, skewed distributions of study-level outcomes θi were identified by histograms and were normalized by transformation to the logarithmic scale:
(7)where Θi is a log-transformed (base 10) study-level outcome that we reported in the study as either 
, 
or 
. The corresponding standard error se(Θi) was approximated as:
(8)Log transformation was applied as shown for Amech, Rmech and thalf datasets. Outcomes were synthesized on the logarithmic scale, transformed back to the original raw scale and reported.
Analysis of heterogeneity
To quantify the extent of inconsistency, or heterogeneity, present between datasets, we calculated Q, I2 and H2 heterogeneity statistics. Q is a measure of total variation and was calculated as the sum of the weighted squared differences between the study-level means θi and the fixed effect estimate 
:
(9)
where
Q is a χ2 distributed statistic with N−1 degrees of freedom and the corresponding P-value PQ was used to evaluate the null hypothesis that all datasets reported the same effect. For example, PQ>0.05 indicates that outcomes are consistent and describe the same effect, whereas PQ<0.05 indicates that outcomes are inconsistent or heterogeneous. H2 is another heterogeneity metric that is independent of the number of datasets available, and describes the relative excess of Q over the degrees of freedom (d.f.):
(10)I2 is a transformation of H2 that describes the percentage of variance that is due to heterogeneity:
(11)H2 is preferred over I2 for highly heterogeneous data because it has an unbound upper limit; however, I2 is more easily interpretable. Values of Q=0, I2=0 or H2=1 indicate that data are homogeneous.
Heterogeneity and publication bias were assessed using funnel plots and cumulative-study exclusion plots. The homogeneity threshold TH was calculated from cumulative exclusion analysis and specifies the percentage of studies that need to be removed (according to maximal Q-reduction criteria) before a homogenous set of studies is attained, as determined by the P-value PQ corresponding to the Q heterogeneity statistic (N.M. and S.V.K., unpublished).
Meta-analysis
Study-level outcomes were synthesized under the assumptions of a random effects model to obtain an overall outcome 
:
(12)where the random effects study-level weights wi were estimated as
(13)and the interstudy variance was approximated using the DerSimonian–Laird estimator:
(14)
where
Q is the heterogeneity statistic (introduced above), c is a scaling factor and N is the number of datasets being synthesized. The standard error corresponding to the overall outcome was estimated as:
(15)and the corresponding confidence intervals were constructed using critical values z1−α/2 obtained from a z-distribution:
(16)where α=0.05 corresponds to a 95% significance level.
Subgroup analyses were conducted to identify sources of heterogeneity and explore the influence of different experimental and biological factors on the outcome of interest. Study-level data were stratified into characteristic groups defined by study-level covariates and study-level outcomes within each subgroup were synthesized as above. When multiple subgroups comparisons were made, the Bonferroni correction was used to adjust the significance threshold to control for false-positive findings:
(17)where α* is the adjusted significance threshold to attain intended error rates α for m subgroup comparisons.
Meta-regression
To assess the bivariate relationship between study-level predictors and outcomes, a random-effects meta-regression model was constructed in the following form:
(18)where β0 is the intercept, βinter is the slope coefficient describing the relationship between predictor xi and outcome θi, ɛi is the intrastudy variability approximated by 
and ηi is the interstudy variability approximated by 
. In addition to the meta-regression analysis conducted to assess the between-study relationship of study-level predictors and outcomes, we also conducted an intrastudy regression analysis to assess the within-study relationship between study-level predictors and outcomes (N.M. and S.V.K., unpublished). Within-study regression coefficients were computed and pooled using the meta-analytic methods described above to estimate an overall effect βintra. The magnitude and sign of βinter and βintra were then compared to evaluate whether the observed relationships between and within studies were in agreement.
Analysis of ATP release mechanisms
To evaluate the cell type-specific mechanisms of ATP release, pharmacological and genetic intervention outcomes were grouped by common molecular targets and their inhibitory effects were estimated and synthesized as above. Owing to the non-specific effects of certain inhibitors, differential pharmacological effects were considered to differentiate between ATP release pathways that shared common pharmacological profiles. Depending on the panel of available inhibitor data for each cell type, more targeted interventions allowed us to narrow down the pathways affected by non-specific inhibitors. For example, carbenoxolone inhibits connexin and pannexin activity, while flufenamic acid is a more targeted connexin inhibitor. For cells in which carbenoxolone inhibited ATP release, but flufenamic acid did not, we inferred the involvement of pannexins. In some cases, non-specific interventions were used to support or dismiss the involvement of several pathways at a time. For instance, cells in which Gd3+ had no effect on ATP release allowed us to putatively dismiss ATP release pathways related to MACs, P2X7, Piezo1 and transient receptor potential (TRP) channels. To reflect the varying degrees of evidence (i.e. availability of data, specificity of interventions, separate replication of findings), we distinguished between cell type-specific mechanisms that were reported once, those that were verified by different methods/groups, and those for which data were available but inconclusive (often due to the non-specific nature of interventions). These outcomes were summarized in a cell-by-mechanism involvement matrix, indicating whether the pathway was involved in ATP release and the degree of evidence supporting the conclusion. Based on the mechanisms summarized in the involvement matrix, a co-occurrence analysis was then conducted to screen for possibly co-dependent pathways of ATP release. Results were summarized in a co-occurrence matrix for cases in which at least 3 cell types could be evaluated, and were expressed as the percentage of cell types in which the pairs of mechanisms were co-occurrent.
To determine whether the separate routes of ATP release were additive, the effects of ATP release inhibitors were used to approximate of relative contribution (RC) of that pathway to ATP release:
(19)The joint contribution (JC) was then computed as:
(20)and the standard error was
(21)95% CI for joint contributions were constructed using Eqn 16. Joint contributions that were insignificantly different from 100% (contained by 95% CI) were interpreted as additive, and those exceeding 100% were interpreted as synergistic. Cell types in which at least 4 of the 5 main routes of ATP release were studied were included in this analysis, except for cell types that exhibited synergy-like interactions, in which case no restriction was placed on the minimum number of routes studied, because including the relative contributions of additional release routes would have only added to the joint contribution.
Footnotes
Competing interests
The research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Author contributions
Conceptualization: N.M., S.V.K.; Methodology: N.M., M.M., S.V.K.; Validation: N.M., A.M., M.M.; Formal analysis: N.M., A.M., S.V.K.; Investigation: N.M.; Resources: M.M.; Data curation: N.M., A.M., S.V.K.; Writing - original draft: N.M., S.V.K.; Writing - review & editing: N.M., A.M., M.M., S.V.K.; Visualization: N.M., S.V.K.; Supervision: S.V.K.; Project administration: S.V.K.; Funding acquisition: S.V.K.
Funding
This work was supported by the Natural Sciences and Engineering Research Council of Canada [RGPIN-288253] and the Canadian Institutes of Health Research [MOP-77643]. N.M. was supported by the Faculty of Dentistry, McGill University and Réseau de Recherche en Santé Buccodentaire et Osseuse.
Supplementary information
Supplementary information available online at http://jcs.biologists.org/lookup/doi/10.1242/jcs.223354.supplemental
- Received July 31, 2018.
- Accepted October 10, 2018.
- © 2018. Published by The Company of Biologists Ltd
References
Article navigation
Related articles
Cited by...
More in this TOC section
Similar articles
Other journals from The Company of Biologists
First Person interviews
Have you seen our First Person interviews with the early-career first authors of our papers? The authors talk about their work in and out of the lab, the journeys that led them to where they are now and the scientists who inspired them along the way. Recently, we caught up with first authors Maitreyi Rathod, Sushmita Chatterjee, Lotte Vanheer, Rachel Furlong, Pamela Adami, Ruiqi Wang and Chen Yu.
Travelling Fellowship – New imaging approach unveils a bigger picture
Find out how Pamela Imperadore’s Travelling Fellowship grant from The Company of Biologists took her to Germany, where she used new imaging techniques to investigate the cellular machinery underlying octopus arm regeneration. Don’t miss the next application deadline for 2020 travel, coming up on 29 November. Where will your research take you?
preLights – Meet the preLighters: an interview with Maiko Kitaoka
Maiko Kitaoka is a graduate student in the lab of Rebecca Heald at the University of California, Berkeley. Here she studies the cause of chromosome mis-segregation defects in Xenopus hybrids. We caught up with Maiko to discuss her research, science communication, ballet, preprints and more.
Journal Meeting – Cell Dynamics: Host-Pathogen Interface
Registration is now open for the third instalment of the highly successful Cellular Dynamics Meeting Series, and will focus on ‘Host-Pathogen Interface’. The meeting will take place 17-20 May 2020, and further information is available here.
Cell Science at a Glance – Adaptor protein complexes and disease at a glance
A new poster from the Robinson lab summarises what is known about the five adaptor protein complexes and discuss how this helps to explain the clinical features of different genetic disorders.
JCS joins the Review Commons initiative
Journal of Cell Science is pleased to be a part of the new and exciting Review Commons initiative, launched by EMBO and ASAPbio. Streamlining the publishing process, Review Commons enables high-quality peer review to take place before journal submission. Papers submitted to Review Commons will be assessed independently of any journal, focusing solely on the paper’s scientific rigor and merit.
Articles of interest in our sister journals
Casein kinase 1α decreases β-catenin levels at adherens junctions to facilitate wound closure in Drosophila larvae
Chang-Ru Tsai, Michael J. Galko
Development
Spherical spindle shape promotes perpendicular cortical orientation by preventing isometric cortical pulling on both spindle poles during C. elegans female meiosis
Elizabeth Vargas, Karen P. McNally, Daniel B. Cortes, Michelle T. Panzica, Brennan M. Danlasky, Qianyan Li, Amy Shaub Maddox, Francis J. McNally
Development